[vbox-dev] Parallel build failure
Malte Starostik
lists at malte.homeip.net
Wed Jul 27 12:38:52 PDT 2011
> > do you know which version kmk gentoo is using for building? Sounds like
> > it's not the same as we are using since we're constantly building
> > VirtualBox on multi-core/cpu boxes and are not seeing this issue.
>
> the Gentoo package is labelled as 0.5.5-r1 (r1 being the package revision):
>
> kmk - kBuild version 0.1.98 (r2274)
>
> Based on GNU Make 3.81.90:
>
> I can try some build cycles with 0.1.5_p2 tomorrow and see if that changes
> anything.
Sorry, took a little longer to collect these results. I tried again with this
version:
kmk - kBuild version 0.1.5 (r2373)
Based on GNU Make 3.81.90:
and VirtualBox 4.1.0.
Testing results:
* All combinations of kmk 0.1.98 resp.0.1.5 and VirtualBox 4.0.12 resp. 4.1.0
exhibit the same spurious build failures.
* Although the error hit me before during normal building (for VB 4.0.10 and
4.0.12) - and both times it took more than one repetition and then going back
to -j1 to get through the build - it wasn't all that easy to reproduce it as
often.
* My normal builds were passing -j3 -l3 to make/kmk. In fact I was about to
raise the job number on this machine, but hadn't done so yet, and this
apparently made me select the magic values - at least for this box:
With -j3 -l3, roughly 40% of the builds fail, all of them with a missing
state/cr_statefuncs.h.
I then tried with -j1, to -j4, and varying values for -l, including no -l
option at all. All of those combinations succeeded with a 100% rate for at
least 10 build runs each.
All in all, this still seems like a make dependency issue to me, although I
think it's rather weird how reproducible only one single combination of make
options makes it bite - especially as the machine was subject to varying
amounts of non-related load in the meantime.
HTH,
Malte
More information about the vbox-dev
mailing list